| Summary: | Using the 1948-2004 NCEP/NCAR reanalysis dataset, an objective ensemble natural analog system is developed to produce probabilistic and categorical 1-4 week forecasts of 1000-500 hPa thickness for the Northern Hemisphere. Natural analog ensembles of 200-500 members are selected by comparing analogous patterns over the Northern Hemisphere from 1948-2003. From the analog ensemble, 1-4 week NH 1000-500 hPa thickness forecasts are produced and verified. Categorical forecasts produced from three equally probable classes (above, normal, below) and verified using the Heidke Skill Score (HSS) are found to be more skillful than random forecasts (climatology) throughout the NH for week 1 and week 2 thickness forecasts. Week 3 and week 4 thickness forecasts are more skillful than climatology throughout all seasons and in all locations in the tropics, while positive skill is found in localized regions in the middle and higher latitudes, especially during winter. Additional experimentation reveals that multivariate pattern matching is potentially useful for development of specialized analog forecast systems. Specifically, comparing historical analogs of 300 hPa winds and 1000-500 hPa thickness provide improvements in week 1-4 forecasts for the central U.S. Furthermore, analog forecasts derived from matching sea surface temperatures and thicknesses are found to enhance skill in regions with documented connections to the El Niño Southern Oscillation (ENSO) (e.g. Western Canada and Alaska, eastern and central Pacific). Augmenting the original thickness analogs with historical sunspot analogs exposes the importance of the solar cycle on the North Atlantic and Polar climate, improving analog thickness forecasts in those regions. Skill of the analog forecasts is observed to generally improve during ENSO events and when patterns in the mid and high latitude are highly anomalous. Moreover, smaller errors in the analog analysis correlate to improvements in week 1-4 NH forecasts, and may provide an a priori estimate of the reliability of the forecasts. Additional datasets and a continually growing archive of historical analyses will undoubtedly improve analog forecasting. Future research in analog forecasting should attempt to exhaust the potential predictability from the limited historical data archive. The parameter space of the proposed analog system presented has not been fully explored, allowing future research to investigate more optimal analog match configurations.
|
|---|